Benchmarking The Ubuntu "Low-Jitter" Linux Kernel

There's an independently maintained "low-jitter" version of the Linux kernel targeting Ubuntu, which claims to be faster, but is that really the case?

In response to the recent Linux 3.7 + Mesa 9.1-devel Running On Ubuntu 12.10 article, a Phoronix reader was quick to promote his specially-configured kernel for Ubuntu.

Said across several messages within the forums, "[Doom 3] runs really really well on my low-jitter kernel, with the closed source drivers though... You can play doom3 at 36 fps? Not really. You can get an idea of it, but the exhilaration of low-jitter 72fps it will not have...Even my mac mini from 2010 plays HL2 ep2 well, and that is much newer. Thats an nvidia 320m processor. (2.4ghz core2duo). Carmack says doom 3 still stresses a lot of hardware, because of the 3 passes he uses to render stuff. Which is also why I linked to my low-jitter kernel, because it runs doom 3 well. (Better than windows)."

The Linux user and Phoronix reader, Uwaysi Bin Kareem, describes his low-jitter kernel for Ubuntu 12.04 LTS as "A lot of Linux-distros use the standard config, or close. Unfortunately the standard config, is neither specialised for desktop, or server. It is somewhere between, a kind of political standpoint, not siding with either camp. Also debugging is usually left on, and low-latency kernel-variants of ubuntu, still are not agressively tuned towards low os-jitter. What that means is that these distros will perform much poorer on the desktop than it can. And that includes the most popular distro “Ubuntu” which I use. However compiling ones own kernel for it, really gets it to the performance one would want."

To see whether this third-party low-jitter kernel for Ubuntu Linux is any faster, I ran some of the common Phoronix Test Suite tests with OpenBenchmarking.org integration on his Linux 3.5.4 low-jitter kernel and compared it to the mainline Linux 3.5.4 kernel as offered from the Ubuntu Mainline PPA. Benchmarking was from an Intel Core i7 laptop with NVIDIA graphics and using the proprietary driver.

Benchmark results in full along with all of the system details can be found in OpenBenchmarking.org's 1210149-RA-LOWJITTER02 result file. Below are just a couple of the interesting results.

For most of the Linux benchmarks between these two kernels, there isn't much to see...

In some tests, the low-jitter kernel is actually a bit slower than the generic/mainline kernel from the 3.5.4 code-base.

It isn't particularly exciting.

Of the many benchmarks run, the only big win for the Ubuntu low-jitter kernel was found when running the PostgreSQL database server.

However, low-jitter isn't really good for all server workloads since it sharply loses out in the Apache web server benchmark.

Continue to OpenBenchmarking.org for the rest of the Linux kernel benchmarks plus other kernel data.